This paper is in the following e-collection/theme issue:

Clinical Information and Decision Making (1373) Electronic Health Records (1031) Machine Learning (1433) Intensive Care Unit (ICU) (109)

Published on in Vol 22, No 3 (2020): March

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/16374, first published .
Learning Latent Space Representations to Predict Patient Outcomes: Model Development and Validation

Learning Latent Space Representations to Predict Patient Outcomes: Model Development and Validation

Learning Latent Space Representations to Predict Patient Outcomes: Model Development and Validation

Authors of this article:

Subendhu Rongali1 Author Orcid Image ;   Adam J Rose2 Author Orcid Image ;   David D McManus3 Author Orcid Image ;   Adarsha S Bajracharya3 Author Orcid Image ;   Alok Kapoor3, 4 Author Orcid Image ;   Edgard Granillo5 Author Orcid Image ;   Hong Yu1, 3, 6, 7 Author Orcid Image
Article Authors Cited by (13) Tweetations (3) Metrics

    Original Paper

    1College of Information and Computer Sciences, University of Massachusetts Amherst, Amherst, MA, United States

    2Section of General Internal Medicine, Boston University School of Medicine, Boston, MA, United States

    3Department of Medicine, University of Massachusetts Medical School, Worcester, MA, United States

    4Meyers Primary Care Institute, Worcester, MA, United States

    5Department of Population and Quantitative Health Sciences, University of Massachusetts Medical School, Worcester, MA, United States

    6Department of Computer Science, University of Massachusetts Lowell, Lowell, MA, United States

    7Center for Healthcare Organization and Implementation Research, Bedford Veterans Affairs Medical Center, Bedford, MA, United States

    Corresponding Author:

    Hong Yu, MA, MS, PhD

    Department of Computer Science

    University of Massachusetts Lowell

    1 University Ave

    Lowell, MA, 01854

    United States

    Phone: 1 978 934 3620

    Email: Hong_Yu@uml.edu


    Background: Scalable and accurate health outcome prediction using electronic health record (EHR) data has gained much attention in research recently. Previous machine learning models mostly ignore relations between different types of clinical data (ie, laboratory components, International Classification of Diseases codes, and medications).

    Objective: This study aimed to model such relations and build predictive models using the EHR data from intensive care units. We developed innovative neural network models and compared them with the widely used logistic regression model and other state-of-the-art neural network models to predict the patient’s mortality using their longitudinal EHR data.

    Methods: We built a set of neural network models that we collectively called as long short-term memory (LSTM) outcome prediction using comprehensive feature relations or in short, CLOUT. Our CLOUT models use a correlational neural network model to identify a latent space representation between different types of discrete clinical features during a patient’s encounter and integrate the latent representation into an LSTM-based predictive model framework. In addition, we designed an ablation experiment to identify risk factors from our CLOUT models. Using physicians’ input as the gold standard, we compared the risk factors identified by both CLOUT and logistic regression models.

    Results: Experiments on the Medical Information Mart for Intensive Care-III dataset (selected patient population: 7537) show that CLOUT (area under the receiver operating characteristic curve=0.89) has surpassed logistic regression (0.82) and other baseline NN models (<0.86). In addition, physicians’ agreement with the CLOUT-derived risk factor rankings was statistically significantly higher than the agreement with the logistic regression model.

    Conclusions: Our results support the applicability of CLOUT for real-world clinical use in identifying patients at high risk of mortality.

    J Med Internet Res 2020;22(3):e16374

    doi:10.2196/16374

    Keywords

    predictive modeling; neural networks; ablation; patient mortality 


    Background

    High-precision predictive modeling of clinical outcomes (eg, adverse events such as the onset of disease and death) is a clinically important but computationally challenging task. If physicians can be notified about the risks of adverse events in advance, they may be able to take steps to prevent them. Electronic health records (EHRs) are widely used in US hospitals and are becoming more mature over time [1]. They have been actively researched for predictive modeling [2-8].

    Almost 6 million patients are admitted annually to intensive care units (ICUs) in the United States for airway support, for hemodynamic or respiratory monitoring, and to stabilize acute or life-threatening medical problems [9-15]. Patients in ICUs are vulnerable to many acute diseases and often suffer from chronic illness, but the leading causes of death in the ICU are multi-organ failure, sepsis, and cardiovascular disease. Approximately 10% to 30% of adult patients die before hospital discharge in ICUs [16-30]. Regression models have been widely used for predicting mortality for ICU patients [31]. Goal-directed sepsis care represents an example of a successful, evidence-based approach to the care of critically ill patients with sepsis that uses predictive modeling to target patients at high risk for mortality with life-saving upstream therapies [21].

    During the past several years, neural network models have shown a great success for many artificial intelligence applications including computer vision, natural language processing, and predictive modeling [4,32-34]. Neural network-based predictive models include the convolutional neural network (CNN) and recurrent neural network (RNN) framework.

    Although studies show that CNN models do not necessarily outperform conventional predictive models such as regression models [35], RNNs [36] have been shown to work well with sequential data such as longitudinal EHRs. There have been promising results regarding the use of RNNs in clinical applications such as diagnosis predictions [6,37,38]. Autoencoders [39] are another class of neural networks that extract rich representations using large unlabeled EHR data and have shown state-of-the-art performance in prediction [4].

    Although NN-based predictive models have been developed, most models are based on bag of features, and few have explicitly modeled the complex relationships between different types of EHR data. Clinical events and diagnoses are not isolated but instead are complex, multifaceted, and often correlated. For example, diagnostic testing leads to a new finding, which may lead to a specific treatment. Therefore, we believe it is important to account for such relationships to improve the predictive power of a model.

    Objective

    The main objective of this work is to develop innovative prediction models to accurately predict patient mortality using patients’ longitudinal EHR data. An important component of our models is a correlational neural network, which is a special neural network model that accounts for correlations between different types of features. We modeled the relationships between different types of clinical features in the EHR through a correlational neural network and integrated them into LSTM-based predictive models for improved performance.

    Contributions

    Our main contributions include learning of latent features from different clinical data types and integrating the learned latent features for outcome prediction using longitudinal EHR data. Our results show that the integration of latent features yielded the highest results for predicting patient mortality using the ICU data.

    In addition to evaluating our CLOUT models using the traditional evaluation metrics such as sensitivity, specificity, and area under the receiver operating characteristic curve, we studied the interpretability of our predictive models. Specifically, we designed a simple ablation experiment [40] to identify important features (or risk factors). Our evaluation results show that physicians were more in agreement with the risk factors ranked by CLOUT than the ones ranked by the commonly used logistic regression model.

    In summary, our contributions are twofold: (1) We developed an innovative long short-term memory (LSTM)–based predictive model where a correlational neural network is integrated to identify relationships and latent representations of different clinical features. Our CLOUT model has state-of-the-art performance in mortality prediction, surpassing other competitive NN models and a logistic regression model. (2) We provide a comprehensive evaluation of risk factors identified by our neural network models. Our results show that the risk factors identified by the CLOUT model agree with physicians’ assessment, suggesting that CLOUT could be used in real-world clinical settings.


    The Medical Information Mart for Intensive Care-III Dataset

    All models are trained and evaluated on the Medical Information Mart for Intensive Care-III (MIMIC-III) dataset; an EHR dataset made publicly available by the Massachusetts Institute of Technology Laboratory for Computational Physiology. MIMIC-III has been widely used for predictive models [41]. The dataset contains 7537 patients with two or more encounters, which is the subset we used to build our CLOUT and baseline models. We call this dataset p-MIMIC. Some demographic information for patients in this dataset is given in Table 1.

    Table 1. Patient demographic information (N=7537).
    CharacteristicValues
    Age (years)

    		Mean74.74

    		Median66.00
    Sex, n (%)

    		Male4190 (55.59)

    		Female3347 (44.41)
    Race, n (%)

    		White5644 (74.88)

    		Black867 (11.50)

    		Hispanic277 (3.68)

    		Asian226 (3.00)

    		Other/unknown523 (6.94)

    We require two or more encounters because we remove the last encounter while making predictions, requiring us to have at least one other encounter with data. We use patient mortality as our outcome label. This label is obtained in the MIMIC dataset from the hospital records and the social security death records. In our dataset of 7537 patients, we have 2825 (37.9%) documented deaths. Further details about MIMIC are covered in Multimedia Appendix 1.

    The dataset was further divided into train, validation, and test splits, each containing approximately 69.99% (5275/7537), 9.99% (753/7537), and 20.02% (1509/7537) of the patients, respectively. Once we picked the optimal model hyper-parameters using the validation set, the model was retrained on the combined train-validation set, which contained 79.98% (6028/7537) of the data.

    Baselines—Reverse Time Attention Model, Time-Aware Reverse Time Attention Model, Logistic Regression Models

    Our first set of baseline models are versions of the RETAIN model, which is one of the few publicly available predictive models for EHRs. RETAIN was built on an RNN model, and evaluation has shown that it achieved both state-of-the-art performance and interpretability [6].

    RETAIN by itself does not incorporate temporal information beyond the RNN framework; such fine-grained temporal information may be important to patient outcomes. For example, the severity of 2 acute myocardial infarctions separated by different durations could have different clinical implications. On the other hand, there is an option to include the time features to the encounter vector. Therefore, we implemented time-aware RETAIN (TaRETAIN) models as additional baselines by concatenating time information to the input features. We experimented with two different approaches to create the time feature: number of days elapsed since the first encounter and number of days elapsed since the previous encounter. We call these 2 models TaRETAIN-first and TaRETAIN-previous.

    Another baseline model is logistic regression as it has been commonly used with EHR data. Although logistic regression is best in interpretability, it is difficult to incorporate temporal information. We therefore combined all the information documented in an encounter to form 1 feature vector for each patient. Our logistic regression model was also augmented with the l2 penalty.

    The CLOUT Models

    The CLOUT models are built upon the state-of-the-art LSTM framework. We provide a description of relevant concepts or components that are built into our CLOUT models in Multimedia Appendix 2.

    Unlike other RNN models, LSTMs can learn dependencies over longer intervals more efficiently [42]. In this study, CLOUT represented all LSTM-based predictive models we built for EHRs. The central architecture, as shown in Figure 1, is an attention-based LSTM model that processed the encounter vectors and made a binary class prediction.

    Given a patient with encounters, the encounter vectors derived from a CLOUT model are e1, e2, ... en. We ran the encounter vectors through the LSTM framework to get the hidden vectors at each time step, h1, h2, ... hn. We then used the attention module to find the weighted sum of these hidden vectors . Formally, H = a1 . h1 + a2 . h2 + ... an . hn. The vector was then sent through a linear layer, and the output was squashed between 0 and 1 using a sigmoid function. This final output represented the probability of a positive class, which in our current application was the probability that the patient died.

    Note that our LSTM architecture is commonly used for sequence data. The innovation of this work is the representation of the encounter vector that integrates different types of EHR data, which we will describe below.

    ‎
    Figure 1. Our model architecture. LSTM: long short-term memory.
    View this figure
    A Simple Concatenation Model

    In this version of CLOUT, the encounter vector was derived by a simple concatenation of different types of features. Every patient encounter had a set of documented International Classification of Diseases (ICD) codes, medications, and laboratory components. We converted these to 3 bit-vectors, , , and , respectively, each of the size of the vocabularies. Bit-vectors are vectors of size equal to the length of the vocabulary with 1 at the index where the feature is documented and 0 everywhere else. We passed these bit-vectors through linear embedding layers to get their dense vector representations. We concatenated these dense vector representations and passed the resultant vector through a nonlinear function such as the rectified linear unit [43] to get the final encounter representation, .

    Representation Through Concatenation With Autoencoders

    Recent work on word embeddings called ELMo [44] has shown that integrating different levels of representations learned by neural networks improves predictive performance in natural language processing applications, as different layers represent different characteristics of input data. Building on the same concept, we created a CLOUT model that integrates the representations of input features learned from an autoencoder with our inputs before sending them through the prediction layer. The hidden layer representations contain valuable information about the relationships between different input features, and by including this information along with the actual input features, we enable the model to make predictions with more knowledge. We integrate the representations using concatenation.

    The Latent Space Representation

    ICD codes, medications, and laboratory results are not isolated unrelated clinical information. They are clinically intertwined or correlated. For example, as stated earlier, medications depend upon the diagnoses of the patient in that encounter. To capture the correlations among EHR data, we added a multi-view latent space component, as shown in Figure 2, by adapting a correlational neural network [45] framework.

    We used a correlational neural network for 3 views (ICD codes, medications, and laboratory components) to construct the latent representation for our latent space model. This component is graphically shown in Figure 3.

    The latent space representation is a measure of the patient condition—a combination of related information from diagnosis codes, medications, and laboratory components. The details of this component are further described in Multimedia Appendix 3.

    To integrate latent space representation into the encounter vector, we first projected the encounter into this latent space to get the latent space vector, l. We simultaneously performed all of the operations in the simple concatenation version to find the encounter vector of that version, ec. The final encounter vector was the concatenation of l and ec. The model described here is shown in Figure 2. Note that the c-operation stands for concatenation.

    To evaluate the effectiveness of the correlational neural network, we also implemented a traditional autoencoder with one hidden layer f and one output layer g with the goal to reconstruct the input using a hidden representation of lower dimensions. We called this model CLOUT-autoencoder.

    ‎
    Figure 2. Model for constructing the encounter vector. ReLU: rectified linear unit; ICD: International Classification of Diseases.
    View this figure
    ‎
    Figure 3. The correlational neural network for our 3 views. ICD: International Classification of Diseases.
    View this figure
    Evaluation

    We evaluated each of the baseline and CLOUT models on the p-MIMIC dataset. We obtain true-positives (TP), false-positives (FP), true-negatives (TN), and false-negatives (FN). We report area under the receiver operating characteristic curve (AUC-ROC) scores for all models, and precision , recall , and F1-scores for the top performing models.

    Risk Factor Experiment With Physicians

    Predictive models would be of limited clinical use if the models are not interpretable. To interpret or identify the risk factors in our CLOUT models, we conduct an ablation experiment, which has been widely used for feature engineering. We perturb the patient data to zero out the contribution of a feature and calculate the corresponding difference in output. This classical method shows the contribution of each feature, which may correspond to the risk score.

    Recall that each of our CLOUT models outputs a probability score that indicates mortality risk. So, the difference in output would be the reduction in this probability, which we call the attribution weight of the given feature. We calculated the attribution weight for each ICD code, medication, and laboratory component that is documented in the patient's EHRs. These features would then constitute the risk factors associated with the mortality, and the attribution weight represents the strength of the association.

    Although ablation experiments have been widely used for feature engineering and interpretation of neural network models in many applications [46], they have not been evaluated for identifying risk factors of patient outcome based on longitudinal EHRs.

    Therefore, we designed a comprehensive evaluation of the risk factors ranked by CLOUT and compared them with ones ranked by a logistic regression model. Specifically, we ranked the risk factors at the patient level and population level. At the patient level, each risk factor (ie, feature or variable) is weighted by its contribution to the correct prediction to the patient. We ranked the risk factors at the population level by aggregating and normalizing the attribution weights of features across the patient population.

    Experiment Design

    Using stratified random sampling, we selected a subset of risk factors from the prediction models CLOUT and logistic regression, respectively, and asked 5 unbiased physicians (4 internists and 1 cardiologist), who were not privy to the reasons for doing the ranking, to independently judge the clinical relevance of those risk factors.

    To reduce the total number of features that the physicians need to evaluate, we selected features from CLOUT. Specifically, for each feature, the ablation experiment output a relevance score. We bin the features into 3 groups: (1) top 20 features, (2) 20-50 features, and (3) the remaining features. From each bin, we randomly selected 4 features. We then randomly selected 1 patient and accordingly obtained a total of 12 features for that patient. We also obtained the ranked list of features by population and followed a similar bin strategy to select another 18 features distributed across the different feature sets (we purposely selected those features that differ from the features we selected from the sample patient so that we could maximize our evaluation features). Therefore, we selected a total of 30 features (12 by a patient and 18 by the population).

    We randomized those 30 features and asked the 5 physicians who are blinded to the CLOUT rankings to evaluate, for each feature, its clinical relevance. Specifically, we asked each physician to score the feature (1-5, with 1 as the least relevant and 5 the most relevant) based on their clinical knowledge or guidelines.

    We calculated the Pearson correlation coefficient between physicians’ scores for pairwise agreements between physicians, and between the CLOUT scores and physicians’ scores. We also performed a t test for statistical significance. We used the same 30 features to evaluate the logistic regression model and, in this case, using the weight assigned by the logistic regression model for the ranking.

    Finally, we performed another evaluation where we first averaged the scores of all the physicians to obtain a representative gold standard. We then computed the correlation coefficients between these scores and the scores from our models and the logistic regression baseline.


    Model Performance

    During our experiments, we found that models using abnormal laboratory components as input (ie, binary coding of normal/abnormal) performed better than those using all the laboratory components. Therefore, the results presented here for the p-MIMIC dataset used only the abnormal labs recorded in patient encounters through an abnormal flag.

    As shown in Table 2, the AUC-ROC results for our CLOUT models are significantly better (P<.001) than both the RETAIN and the logistic regression models. The AUC-ROC curves for the representative models are presented in Figure 4. Our CLOUT model with concatenated latent space representation (Figure 2) achieved 0.89 AUC-ROC score, which is more than 0.06 absolute increase over the ICD-RETAIN, logistic regression, and simple LSTM models and a 0.02 increase over RETAIN with all codes. To get a better understanding of our results, we also present the precision and recall scores for each class for the top models in Table 3.

    Our latent space representation model also slightly outperformed the traditional autoencoder CLOUT model, although it is not statistically significant. An important result here is the integration of different levels of representations (input space and from either an autoencoder or a correlational autoencoder) substantially improves the performance of a model, which outperforms one that uses autoencoder alone. The code for our models and experiments can be found at our CLOUT repository [47].

    Table 2. Area under the receiver operating characteristic curve scores for different models.
    MethodArea under the receiver operating characteristic curve, mean (SD)
    Logistic regression0.82 (0.0103)
    RETAINa (only ICDb)0.82 (0.0924)
    TaRETAINc-first (only ICD)0.82 (0.0118)
    TaRETAIN-prev (only ICD)0.82 (0.0919)
    RETAIN (all codes)0.86 (0.0105)
    Long short-term memory with only ICD codes0.83 (0.0104)
    CLOUTd—only autoencoder0.80 (0.0116)
    CLOUT—only latent space0.81 (0.0082)
    CLOUT—simple concatenation0.88 (0.0096)
    CLOUT—autoencoder concatenation0.88 (0.0107)
    CLOUT—latent space concatenation0.89 (0.0138)e

    aRETAIN: Reverse Time Attention model.

    bICD: International Classification of Diseases.

    cTaRETAIN: time-aware RETAIN.

    dCLOUT: L(STM) Outcome prediction using Comprehensive features relations.

    eBest performing model.

    ‎
    Figure 4. The area under the receiver operating characteristic curves for various models. RETAIN: Reverse Time Attention model; CLOUT: L(STM) Outcome prediction using Comprehensive feature relations.
    View this figure
    Table 3. Precision, recall, and F-scores for top CLOUTa models.
    Method and classPrecisionRecallF-score
    CLOUT—Simple concatenation

    		00.850.820.83

    		10.710.760.73

    		Average0.800.790.80
    CLOUT—Autoencoder concatenation

    		00.850.850.85

    		10.740.740.74

    		Average0.810.810.81
    CLOUT—Latent space concatenation

    		00.840.880.86

    		10.780.720.72

    		Average0.820.820.82

    aCLOUT: L(STM) Outcome prediction using Comprehensive features relations.

    Risk Factors

    To measure agreements among physicians, we compute the Pearson correlation coefficient between their scores. For patient-specific features, Table 4 shows the Pearson correlation coefficient between each pair of physicians and also between different models and the physicians. With the physician gold standard ratings computed by averaging, we found that our model had a correlation coefficient of 0.64, which is higher (4.9%) than the correlation coefficient of 0.61 with the logistic regression model.

    Table 4. Pearson correlation coefficients for agreement between physicians and models.
    AgreementPhysician 1, rPhysician 2, rPhysician 3, rPhysician 4, rPhysician 5, rMean (SD)
    Physician-physician agreement

    		Physician 11.000.810.560.610.880.72 (0.13)

    		Physician 20.811.000.870.650.860.80 (0.09)

    		Physician 30.560.871.000.490.690.65 (0.14)

    		Physician 40.610.650.491.000.610.59 (0.06)

    		Physician 50.880.860.690.611.000.76 (0.11)
    Physician-model agreement

    		Logistic regression0.600.630.530.320.520.52 (0.11)

    		RETAINa0.650.720.610.300.580.57 (0.14)

    		CLOUTb—only autoencoder−0.070.130.210.55c0.170.20 (0.20)

    		CLOUT—only latent space0.420.770.640.350.530.54 (0.15)

    		CLOUT—simple concatenation0.520.640.700.190.670.54 (0.19)

    		CLOUT—autoencoder concatenation0.540.700.640.140.620.53 (0.20)

    		CLOUT—latent space concatenation0.690.770.590.180.670.58 (0.21)

    aRETAIN: Reverse Time Attention model.

    bCLOUT: L(STM) Outcome prediction using Comprehensive features relations.

    cItalicization signifies highest physician-model agreement in the column.


    Principal Findings

    In this study, we have developed innovative CLOUT models and compared them with other state-of-the-art predictive models with respect to performance on mortality prediction. We found that the performance of almost every CLOUT model surpassed the competitive baseline models (eg, RETAIN). The results support that LSTM is a state-of-the-art framework for EHR-based predictive modeling.

    Our results showed that the integration of different levels of latent representations (input space and from either an autoencoder or a correlational autoencoder) substantially improves the performance from 0.80 to 0.88 AUC-ROC. The rich representation may provide extra information to the model, which in turn helps the model make better predictions. The integration of different types of features (ie, ICD codes, laboratories, and medications) however had a mixed result. Specifically, the CLOUT model that incorporated only the abnormal laboratory results slightly surpassed the CLOUT model that incorporated all 3 features. This supported the importance of laboratory results for predicting mortality. Our results also suggested that there may be noisy information in the features. When CLOUT was implemented with the latent vectors included, it had the highest performance, an AUC-ROC score of 0.89 and an F1 score of 0.82. The result supports our approach of using the correlational neural network to identify latent vectors to best represent different but related clinical observations or variables.

    On the other hand, when we incorporated temporal information as a feature, we showed little improvement in performance using RETAIN. A possible future direction is to explore time-dependent attentions, which may allow the model to integrate the temporal information in the architecture.

    For the risk factors identified by our models, the average correlation coefficient between the physicians is mean 0.71 (SD 0.13), and the average Pearson correlation coefficients between CLOUT and the physicians and between logistic regression and physicians were 0.58 (SD 0.21) and 0.52 (SD 0.11), respectively. These results show a significant difference between the agreement among physicians and the agreement between the logistic regression model and the physicians (P=.04). In contrast, the difference in agreement between the CLOUT models and physicians is not statistically significant, strongly supporting the validity of risk factors and their ranking identified by CLOUT.

    We also calculated the agreement with RETAIN for reference, and we found that the average was 0.57 (SD 0.14), which is still slightly less than the CLOUT model, with CLOUT losing out a lot with physician 4. Other CLOUT models also have slightly lower scores as reported in Table 4, but it is notable that the latent vector models that use the correlational autoencoder have better correlations (0.58, SD 0.21) with physicians than the ones that use a simple autoencoder (0.53, SD 0.20). The evaluation with our gold standard (the average physician scores) also informs us that CLOUT selects more meaningful features.

    Our results show that physician 4 had a low correlation score with other physicians as well as with our CLOUT models. For example, lactulose enema, and encephalopathy not otherwise specified were scored as 2 by physician 4, whereas all the other physicians gave scores of 4 or greater. When we removed physician 4, the correlation between the latent space CLOUT model and the physicians improved from 0.58 to 0.68.

    For population-level features, we performed similar evaluation between physician scores and the CLOUT model scores, and the average correlation coefficient values were ICD codes −0.19, medications −0.43, and laboratory components −0.37, which are lower than patient-specific interpretations. This is not surprising as many risk factors (eg, severe diseases) are rare events that are not present for patients in general.

    Furthermore, CLOUT models captured important risk factors while making predictions. In general, our CLOUT models show that patients with diagnosis codes representing cranial nerve disorder and cystic liver disease were marked with a high risk of mortality. This is reasonable as those are diseases with a high risk of mortality.

    Limitations

    Our dataset was constructed from EHR data and is, hence, prone to standard data quality issues that EHRs typically have, as documented in the literature. EHRs are known to have missing diagnoses and medication codes for patients when compared with insurance claims. Furthermore, our analysis of ICU admissions does not account for death because of accidental circumstances such as car crashes. We used all the information exactly as it appears as it is infeasible to comb through all the records to pick patients for the study. Another limitation we would like to report is the absence of vital sign features in our dataset, which we ignore because of the involved preprocessing steps that are required to handle missing numerical values.

    The CLOUT models have significant limitations as well. First, similar to most predictive models, the risk factors identified by the CLOUT models include cofounding variables. For example, we found that patients who have a prescription for a scopolamine patch have high-risk scores. This is a medication prescribed to terminally ill patients as part of palliative care regimen to reduce excessive airway secretion. So, in this case, the actual reason for palliative care is a strong risk factor for death, not the medication, which is a confounding factor. Another limitation of our work is that our models are very dependent on the population size. Bias could be introduced when the size is small. However, such limitations exist in most predictive models not reviewed or guided by physician oversight.

    Comparison With Prior Work

    We surveyed a variety of approaches to compare our models. This includes statistical approaches [48-52], deep learning–based approaches [6,38,53-55], and other phenotyping efforts [4,5]. We also surveyed papers on interpretability. A detailed analysis of all this can be found in Multimedia Appendix 4.

    Conclusions

    EHRs are widely available and have enormous untapped potential to predict patients’ health outcome. EHR-based predictive models are potentially hugely useful for clinical decision support. Our experiments show that incorporating comprehensive clinical information is useful and can improve predictions and that integrating latent space representations learned through a correlational neural network to clinical information led to the best performing CLOUT model. Our risk factor experiment with physicians also suggests that CLOUT models find more clinically relevant risk factors. Our results strongly support that CLOUT may be a useful tool to generate clinical prediction models, especially among hospitalized and critically ill patient populations.

    The future directions include new models to incorporate the temporal information and methods to integrate clinical notes for predictive models. We may also explore other models to integrate different views, including the Capsule network model [56].

    Acknowledgments

    This work was supported in part by the grant R01HL137794 from the National Institutes of Health (NIH). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. HY’s time was also supported by grants 5R01HL125089, R01HL135129, R01DA045816, and R01LM012817. DM’s time was also supported by grants R01HL137734, R01HL126911, R01HL13660, and R01HL141434 from the National Heart, Lung, and Blood Institute.

    Conflicts of Interest

    DM has received research grant support from Apple Computer, Bristol-Myers Squibb, Boehringher-Ingelheim, Pfizer, Samsung, Philips Healthcare, Care Evolution, and Biotronik; has received consultancy fees from Bristol-Myers Squibb, Pfizer, Flexcon, and Boston Biomedical Associates; and has inventor equity in Mobile Sense Technologies, Inc, Connecticut.

    ‎
    Multimedia Appendix 1

    The Medical Information Mart for Intensive Care-III, preprocessing, and outcome label.

    DOCX File , 16 KB
    ‎
    Multimedia Appendix 2

    Relevant machine learning components.

    DOCX File , 14 KB
    ‎
    Multimedia Appendix 3

    Correlational neural network.

    DOCX File , 15 KB
    ‎
    Multimedia Appendix 4

    Comparison with prior work.

    DOCX File , 17 KB
    1. Kharrazi H, Gonzalez C, Lowe K, Huerta T, Ford E. Forecasting the maturation of electronic health record functions among US hospitals: retrospective analysis and predictive model. J Med Internet Res 2018;20(8):e10458 preprint. [CrossRef]
    2. Jha AK, DesRoches CM, Campbell EG, Donelan K, Rao SR, Ferris TG, et al. Use of electronic health records in US hospitals. N Engl J Med 2009 Apr 16;360(16):1628-1638. [CrossRef] [Medline]
    3. Choi E, Bahadori MT, Schuetz A, Stewart WF, Sun J. Doctor AI: Predicting clinical events via recurrent neural networks. JMLR Workshop Conf Proc 2016 Aug;56:301-318 [FREE Full text] [Medline]
    4. Miotto R, Li L, Kidd BA, Dudley JT. Deep patient: an unsupervised representation to predict the future of patients from the electronic health records. Sci Rep 2016 May 17;6:26094 [FREE Full text] [CrossRef] [Medline]
    5. Zhou J, Wang F, Hu J, Ye J. From Micro to Macro: Data Driven Phenotyping by Densification of Longitudinal Electronic Medical Records. In: Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining.: Association for Computing Machinery, New York, NY, United States; 2014 Presented at: KDD'14; August 24 - 27, 2014; New York, NY, USA p. 135-144. [CrossRef]
    6. Choi E, Bahadori MT, Sun J, Kulas J, Schuetz A, Stewart W. RETAIN: An Interpretable Predictive Model for Healthcare using Reverse Time Attention Mechanism. In: Proceedings of the 2016 Conference on Neural Information Processing Systems. 2016 Presented at: NIPS'16; December 5-10, 2016; Barcelona, Spain p. 3504-3512.
    7. Esteban C, Staeck O, Baier S, Yang Y, Tresp V. Predicting Clinical Events by Combining Static and Dynamic Information Using Recurrent Neural Networks. In: Proceedings of the 2016 IEEE International Conference on Healthcare Informatics.: IEEE; 2016 Presented at: ICHI'16; October 4-7, 2016; Chicago, IL, USA p. 93-101. [CrossRef]
    8. Che Z, Purushotham S, Khemani R, Liu Y. arXiv preprints. 2015. Distilling Knowledge from Deep Networks with Applications to Healthcare Domain   URL: https://arxiv.org/abs/1512.03542 [accessed 2020-03-09]
    9. Wunsch H, Angus DC, Harrison DA, Linde-Zwirble WT, Rowan KM. Comparison of medical admissions to intensive care units in the United States and United Kingdom. Am J Respir Crit Care Med 2011 Jun 15;183(12):1666-1673. [CrossRef] [Medline]
    10. Wunsch H, Wagner J, Herlim M, Chong DH, Kramer AA, Halpern SD. ICU occupancy and mechanical ventilator use in the United States. Crit Care Med 2013 Dec;41(12):2712-2719 [FREE Full text] [CrossRef] [Medline]
    11. Barrett M, Smith M, Elixhauser A, Honigman L, Pines J. NCBI - NIH. 2014. Utilization of Intensive Care Services, 2011: Statistical Brief #185   URL: https://www.ncbi.nlm.nih.gov/pubmed/25654157 [accessed 2020-03-11]
    12. Edwards JD, Houtrow AJ, Vasilevskis EE, Rehm RS, Markovitz BP, Graham RJ, et al. Chronic conditions among children admitted to US pediatric intensive care units: their prevalence and impact on risk for mortality and prolonged length of stay*. Crit Care Med 2012 Jul;40(7):2196-2203 [FREE Full text] [CrossRef] [Medline]
    13. Krmpotic K, Lobos A. Clinical profile of children requiring early unplanned admission to the PICU. Hosp Pediatr 2013 Jul;3(3):212-218 [FREE Full text] [CrossRef] [Medline]
    14. Harrison W, Goodman D. Epidemiologic trends in neonatal intensive care, 2007-2012. JAMA Pediatr 2015 Sep;169(9):855-862. [CrossRef] [Medline]
    15. Pollack MM, Holubkov R, Funai T, Clark A, Berger JT, Meert K, Eunice Kennedy Shriver National Institute of Child Health and Human Development Collaborative Pediatric Critical Care Research Network. Pediatric intensive care outcomes: development of new morbidities during pediatric critical care. Pediatr Crit Care Med 2014 Nov;15(9):821-827 [FREE Full text] [CrossRef] [Medline]
    16. Dombrovskiy VY, Martin AA, Sunderram J, Paz HL. Rapid increase in hospitalization and mortality rates for severe sepsis in the United States: a trend analysis from 1993 to 2003. Crit Care Med 2007 May;35(5):1244-1250. [CrossRef] [Medline]
    17. Druml W, Lenz K, Laggner AN. Our paper 20 years later: from acute renal failure to acute kidney injury--the metamorphosis of a syndrome. Intensive Care Med 2015 Nov;41(11):1941-1949. [CrossRef] [Medline]
    18. Elias KM, Moromizato T, Gibbons FK, Christopher KB. Derivation and validation of the acute organ failure score to predict outcome in critically ill patients: a cohort study. Crit Care Med 2015 Apr;43(4):856-864. [CrossRef] [Medline]
    19. Levy MM, Dellinger RP, Townsend SR, Linde-Zwirble WT, Marshall JC, Bion J, et al. The Surviving Sepsis Campaign: results of an international guideline-based performance improvement program targeting severe sepsis. Intensive Care Med 2010 Feb;36(2):222-231 [FREE Full text] [CrossRef] [Medline]
    20. Randolph AG, McCulloh RJ. Pediatric sepsis: important considerations for diagnosing and managing severe infections in infants, children, and adolescents. Virulence 2014 Jan 1;5(1):179-189 [FREE Full text] [CrossRef] [Medline]
    21. Gupta RG, Hartigan SM, Kashiouris MG, Sessler CN, Bearman GM. Early goal-directed resuscitation of patients with septic shock: current evidence and future directions. Crit Care 2015 Aug 28;19:286 [FREE Full text] [CrossRef] [Medline]
    22. Weiss SL, Fitzgerald JC, Pappachan J, Wheeler D, Jaramillo-Bustamante JC, Salloo A, Sepsis Prevalence‚ Outcomes‚Therapies (SPROUT) Study Investigators and Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) Network. Global epidemiology of pediatric severe sepsis: the sepsis prevalence, outcomes, and therapies study. Am J Respir Crit Care Med 2015 May 15;191(10):1147-1157 [FREE Full text] [CrossRef] [Medline]
    23. Wunsch H, Guerra C, Barnato AE, Angus DC, Li G, Linde-Zwirble WT. Three-year outcomes for Medicare beneficiaries who survive intensive care. J Am Med Assoc 2010 Mar 3;303(9):849-856. [CrossRef] [Medline]
    24. Halpern NA, Pastores SM. Critical care medicine in the United States 2000-2005: an analysis of bed numbers, occupancy rates, payer mix, and costs. Crit Care Med 2010 Jan;38(1):65-71. [CrossRef] [Medline]
    25. Talmor D, Shapiro N, Greenberg D, Stone PW, Neumann PJ. When is critical care medicine cost-effective? A systematic review of the cost-effectiveness literature. Crit Care Med 2006 Nov;34(11):2738-2747. [CrossRef] [Medline]
    26. Banerjee R, Naessens JM, Seferian EG, Gajic O, Moriarty JP, Johnson MG, et al. Economic implications of nighttime attending intensivist coverage in a medical intensive care unit. Crit Care Med 2011 Jun;39(6):1257-1262 [FREE Full text] [CrossRef] [Medline]
    27. Pronovost PJ, Needham DM, Waters H, Birkmeyer CM, Calinawan JR, Birkmeyer JD, et al. Intensive care unit physician staffing: financial modeling of the Leapfrog standard. Crit Care Med 2004 Jun;32(6):1247-1253. [CrossRef] [Medline]
    28. Parikh A, Huang SA, Murthy P, Dombrovskiy V, Nolledo M, Lefton R, et al. Quality improvement and cost savings after implementation of the Leapfrog intensive care unit physician staffing standard at a community teaching hospital. Crit Care Med 2012 Oct;40(10):2754-2759. [CrossRef] [Medline]
    29. Kumar S, Merchant S, Reynolds R. Tele-ICU: efficacy and cost-effectiveness approach of remotely managing the critical care. Open Med Inform J 2013;7:24-29 [FREE Full text] [CrossRef] [Medline]
    30. Young LB, Chan PS, Lu X, Nallamothu BK, Sasson C, Cram PM. Impact of telemedicine intensive care unit coverage on patient outcomes: a systematic review and meta-analysis. Arch Intern Med 2011 Mar 28;171(6):498-506. [CrossRef] [Medline]
    31. Pirracchio R, Petersen ML, Carone M, Rigon MR, Chevret S, van der Laan MJ. Mortality prediction in intensive care units with the Super ICU Learner Algorithm (SICULA): a population-based study. Lancet Respir Med 2015 Jan;3(1):42-52 [FREE Full text] [CrossRef] [Medline]
    32. Chorowski JK, Bahdanau D, Serdyuk D, Cho K, Bengio Y. Attention-Based Models for Speech Recognition. In: Proceedings of the 28th International Conference on Neural Information Processing Systems. 2015 Presented at: NIPS'15; December 2015; Montreal, Quebec, Canada p. 577-585.
    33. LeCun Y, Boser B, Denker JS, Henderson D, Howard RE, Hubbard W, et al. Backpropagation applied to handwritten zip code recognition. Neural Comput 1989;1(4):541-551. [CrossRef]
    34. Cho K, van Merriënboer B, Gulcehre C, Bahdanau D, Bougares F, Schwenk H, et al. Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation. In: Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing.: ACL; 2014 Presented at: EMNLP'14; October 25–29, 2014; Doha, Qatar p. 1724-1734. [CrossRef]
    35. Xie J, Wang Q. Benchmark Machine Learning Approaches with Classical Time Series Approaches on the Blood Glucose Level Prediction Challenge. In: Proceedings of the 2018 International Joint Conference on Artificial Intelligence. 2018 Presented at: IJCAI'18; July 13-19, 2018; Stockholm, Sweden.
    36. Mikolov T, Karafiat M, Burget L, Cernocky J, Khudanpur S. Recurrent Neural Network Based Language Model. In: 11th Annual Conference of the International Speech Communication Association. 2010 Presented at: ISCA'10; September 26-30, 2010; Makuhari, Chiba, Japan.
    37. Lipton ZC, Kale DC, Elkan C, Wetzel R. arXiv preprints. 2015. Learning to Diagnose with LSTM Recurrent Neural Networks   URL: https://arxiv.org/abs/1511.03677 [accessed 2020-03-09]
    38. Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, et al. Scalable and accurate deep learning with electronic health records. NPJ Digit Med 2018;1:18 [FREE Full text] [CrossRef] [Medline]
    39. Vincent P, Larochelle H, Lajoie I, Bengio Y, Manzagol PA. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. J Mach Learn Res 2010;11:3371-3408 [FREE Full text]
    40. Meyes R, Lu M, de Puiseau CW, Meisen T. arXiv preprints. 2019. Ablation Studies in Artificial Neural Networks   URL: https://arxiv.org/abs/1901.08644 [accessed 2020-03-09]
    41. Johnson AE, Pollard TJ, Shen L, Lehman LH, Feng M, Ghassemi M, et al. MIMIC-III, a freely accessible critical care database. Sci Data 2016 May 24;3:160035 [FREE Full text] [CrossRef] [Medline]
    42. Hochreiter S, Schmidhuber J. Long short-term memory. Neural Comput 1997 Nov 15;9(8):1735-1780. [CrossRef] [Medline]
    43. Nair V, Hinton GE. Rectified Linear Units Improve Restricted Boltzmann Machines. In: Proceedings of the 27th International Conference on International Conference on Machine Learning. 2010 Presented at: ICML'10; June 21-24, 2010; Haifa, Israel p. 807-814.
    44. Peters ME, Neumann M, Iyyer M, Gardner M, Clark C, Lee K, et al. Deep Contextualized Word Representations. In: Proceedings of NAACL-HLT 2018. 2018 Presented at: ACL'18; June 1 - 6, 2018; New Orleans, Louisiana, USA p. 2227-2237.
    45. Chandar S, Khapra MM, Larochelle H, Ravindran B. Correlational neural networks. Neural Comput 2016 Feb;28(2):257-285. [CrossRef] [Medline]
    46. Munkhdalai T, Yu H. arXiv preprints. 2016. Neural Semantic Encoders   URL: http://arxiv.org/abs/1607.04315 [accessed 2020-03-09]
    47. GitHub. CLOUT Repository   URL: https://github.com/subendhu19/CLOUT [accessed 2020-03-09]
    48. Harrell Jr FE. Hbiostat. 2014. Regression Modeling Strategies   URL: http://hbiostat.org/doc/rms.pdf [accessed 2020-03-09]
    49. Genders TS, Steyerberg EW, Alkadhi H, Leschka S, Desbiolles L, Nieman K, CAD Consortium. A clinical prediction rule for the diagnosis of coronary artery disease: validation, updating, and extension. Eur Heart J 2011 Jun;32(11):1316-1330. [CrossRef] [Medline]
    50. Binenbaum G, Ying GS, Quinn GE, Dreiseitl S, Karp K, Roberts RS, Premature Infants in Need of Transfusion Study Group. A clinical prediction model to stratify retinopathy of prematurity risk using postnatal weight gain. Pediatrics 2011 Mar;127(3):e607-e614 [FREE Full text] [CrossRef] [Medline]
    51. Tanner L, Schreiber M, Low JG, Ong A, Tolfvenstam T, Lai YL, et al. Decision tree algorithms predict the diagnosis and outcome of dengue fever in the early phase of illness. PLoS Negl Trop Dis 2008 Mar 12;2(3):e196 [FREE Full text] [CrossRef] [Medline]
    52. Pinzón-Sánchez C, Cabrera V, Ruegg P. Decision tree analysis of treatment strategies for mild and moderate cases of clinical mastitis occurring in early lactation. J Dairy Sci 2011 Apr;94(4):1873-1892 [FREE Full text] [CrossRef] [Medline]
    53. Ma F, Chitta R, Zhou J, You Q, Sun T, Gao J. Dipole: Diagnosis Prediction in Healthcare via Attention-based Bidirectional Recurrent Neural Networks. In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2017 Presented at: KDD'17; August 13 - 17, 2017; Halifax, NS, Canada p. 1903-1911. [CrossRef]
    54. Ma T, Xiao C, Wang F. Health-ATM: A Deep Architecture for Multifaceted Patient Health Record Representation and Risk Prediction. In: Health-ATM: A Deep Architecture for Multifaceted Patient Health Record Representation and Risk Prediction. 2018 Presented at: SIAM'18; May 2018; San Diego, California, USA p. 261-269. [CrossRef]
    55. Xiao C, Choi E, Sun J. Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review. J Am Med Inform Assoc 2018 Oct 1;25(10):1419-1428 [FREE Full text] [CrossRef] [Medline]
    56. Sabour S, Frosst N, Hinton GE. Dynamic Routing Between Capsules. In: Proceedings of the 31st Conference on Neural Information Processing Systems. 2017 Presented at: NIPS'17; December 4-9, 2017; Long Beach, CA, USA p. 3586-3866.

    ‎
    CNN: convolutional neural network
    CLOUT: L(STM) Outcome prediction using Comprehensive features relations
    EHR: electronic heath record
    ICD: International Classification of Diseases
    ICU: intensive care unit
    LSTM: long short-term memory
    MIMIC-III: Medical Information Mart for Intensive Care-III
    RETAIN: Reverse Time Attention model
    RNN: recurrent neural network
    TaRETAIN: time-aware RETAIN

    Edited by M Focsa, G Eysenbach; submitted 23.09.19; peer-reviewed by H Kharrazi, A Leichtle; comments to author 18.11.19; revised version received 27.01.20; accepted 14.02.20; published 23.03.20

    Copyright

    ©Subendhu Rongali, Adam J Rose, David D McManus, Adarsha S Bajracharya, Alok Kapoor, Edgard Granillo, Hong Yu. Originally published in the Journal of Medical Internet Research (http://www.jmir.org), 23.03.2020.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.